How to Get Mentioned by ChatGPT: A Tested, Step-by-Step Playbook

The playbook at a glance

1. Record a baseline before changing anything.
2. Fix crawl access for OAI-SearchBot.
3. Audit cited sources in your category to decide between owned pages and earned coverage.
4. Strengthen owned pages with citable substance.
5. Earn third-party mentions on the sites already being cited.
6. Re-test and measure three separate signals: crawler visits, mentions and citations, and referrals.

The order follows dependency. You need a baseline so you can measure change, and crawl access has to come before content work, because content ChatGPT search can't read will never be cited.

Understand the two ways ChatGPT can name your brand

ChatGPT can mention a brand from what the model learned during training, or it can retrieve and cite pages through ChatGPT search. Search is the path you can influence soonest and verify directly. That is why this playbook focuses on it.

The two paths use different crawlers. Blocking or allowing the training crawler has no effect on search. Many sites get this wrong, and the next step explains how to avoid it.

Step 1: Record a baseline before you change anything

A single ChatGPT screenshot proves nothing. ChatGPT builds each answer fresh from the sources it retrieves, so the same question can produce different brands on different runs. To show that a change worked, you need a fixed test you can repeat.

Build a prompt set that covers three kinds of questions:

• Brand prompts: "What is [Brand]?" or "Is [Brand] good for [use case]?"
• Category prompts: "What are the best tools for [job]?" or "How do I [solve problem]?"
• Comparison prompts: "[Brand] vs [Competitor]" or "Alternatives to [Competitor]"

Run each prompt several times and log the results in a sheet like this (illustrative):

Practical guidance, not a research-backed threshold: 20–40 prompts, each run 3–5 times in fresh sessions, gives you a baseline stable enough to compare against later. No published study sets a required number of runs, so keep the prompt set and run count identical between your baseline and re-test.

Keep the "Domains cited" column. It feeds Step 3.

Step 2: Make your site eligible for ChatGPT search

OAI-SearchBot is the robots.txt token that decides whether your site can appear in ChatGPT search answers. OpenAI treats its crawler settings as independent of one another. A site can allow OAI-SearchBot to appear in search while disallowing GPTBot, which governs training use. Sites that opt out of OAI-SearchBot "will not be shown in ChatGPT search answers" (OpenAI).

A minimal configuration that allows search and makes a separate choice about training:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /   # or Allow: / — this choice does not affect search

Then check three things that commonly break eligibility:

• Your firewall or CDN. OpenAI recommends allowing requests from its published IP ranges as well as the robots.txt token (OpenAI). A bot-blocking rule at Cloudflare or a WAF can undo a correct robots.txt.
• Accidental noindex tags. If you want a page hidden, blocking OAI-SearchBot isn't enough. OpenAI may still show the link and title if it learns the URL from a third-party search provider. To hide a page fully, use noindex and leave the page crawlable so the tag can be read (OpenAI Help Center). Check the reverse too: key product pages should not carry a stray noindex.
• Timing. A robots.txt change takes about 24 hours to affect ChatGPT search (OpenAI). Wait at least that long before re-testing a crawl fix. OpenAI gives no timeline for how long new content takes to be cited.

What good looks like: OAI-SearchBot is allowed, OpenAI's IP ranges pass your firewall, and you can see verified OAI-SearchBot requests in your server or edge logs. Eligibility is necessary but not sufficient. OpenAI says allowing the crawler helps ensure you can appear, not that you will.

Step 3: Audit which sources ChatGPT cites in your category

Before you decide between rewriting your site and pitching publishers, look at the "Domains cited" column from your baseline. The research on this question conflicts, and the conflict is the reason to audit your own category rather than follow a generic rule.

The case for earned media. A University of Toronto study using "Top 10 brands" ranking prompts found AI search engines returned 67–92% earned-media citations across automotive, consumer electronics and software, with social platforms nearly absent from ChatGPT and Claude outputs. In US consumer electronics, AI search cited brand sites about 8% of the time, compared with 32.9% on Google (Chen, Wang, Chen & Koudas, 2025, as cited in Capston). Note that "earned" in this study included Wikipedia and government sites.

The case for owned pages. Parse classified 3,805,095 citations that backed a brand recommendation. On ChatGPT Search, the recommended brand's own website supplied 50.4% of that evidence, compared with 19.1% on Google AI Mode (Parse). Own-site share was highest in B2B categories (Professional Services 40.2%, Data & Analytics 37.1%, AI 31.7%) and lowest in consumer ones (Consumer Electronics 10.3%). Commercial media supplied 51.1% in Financial Services and 11.0% in Software. Parse is a vendor, the data is correlational, and 13.2% of citations couldn't be classified.

The studies measured different things: ranking prompts in one case, recommendation-backing citations in the other, with different samples. Taken together, they point to a decision rule:

As a rough prior, B2B and software buyers tend to see owned pages carry more weight, while consumer and financial categories lean on third parties. Your own audit should override that prior.

One common assumption doesn't hold up for ChatGPT: Reddit is not the main lever. Reddit and YouTube were the largest single domains across all engines in Parse's data, but most of those citations came from engines other than ChatGPT Search. Community and social sources made up 5.2% of ChatGPT Search's recommendation evidence (Parse).

Step 4: Rewrite owned pages so they carry citable substance

The strongest peer-reviewed evidence for content changes comes from the GEO study (Aggarwal et al., KDD 2024), which tested content edits across 10,000 queries in 25 domains. Adding citations to sources, quotations and statistics raised a page's visibility in generative answers. Keyword stuffing did not (Aggarwal et al.).

The best results were Quotation Addition (+41%) and Cite Sources (+30%) on a position-adjusted word-count measure, and Statistics Addition (+37%) on a subjective-impression measure (GEO Wiki). These are best cases for particular methods, not averages. The study tested a GPT-3.5 research setup and Perplexity, not ChatGPT Search, so treat the direction as useful and the percentages as non-transferable.

In practice, make each key passage able to stand alone with a claim, a figure and a source. For example (illustrative):

Before: "Our platform is the leading solution for fast, easy invoicing that saves teams time."

After: "Acme sends a first invoice in under two minutes from a blank account. In our March onboarding data, the median new user sent their first invoice 1 minute 48 seconds after signup. Unlike Competitor X, Acme supports recurring invoices on the free plan."

The "after" version gives ChatGPT a specific, attributable claim to quote. The "before" version gives it nothing. Only publish figures you can back up.

Where to apply this first, based on what Parse found backing ChatGPT recommendations:

• Comparison and alternatives pages. Among vendor pages that backed a recommendation, 21.4% had roundup-style titles (Parse). An honest comparison page on your own site is a legitimate candidate for citation.
• Pricing, documentation and use-case pages. These answer the specific questions buyers ask when they compare options.

There's good news for smaller brands. In the GEO study, weaker pages gained the most: Cite Sources lifted pages ranked fifth by 115.1% (GEO Wiki). There's also a caution. A later benchmark found many rewrites ineffective or counterproductive once several parties optimize against the same engine (Puerto et al., NeurIPS 2025, as cited in GEO Wiki). Substance is the durable edge. Formatting tricks are not.

Step 5: Earn mentions on the sites ChatGPT already cites

If your audit shows third-party domains dominating, go to them directly. Your baseline sheet already lists the review sites, roundups and publications ChatGPT cites for your category prompts. Those are your target list.

For each domain:

1. Find the specific cited page, such as a "best tools for X" roundup.
2. Check whether you belong on it by your fit for the use case, not just by category.
3. Pitch the author with something useful: a correction, a data point, access for testing, or a clear explanation of what you do differently.
4. Update your listings on review platforms that appear in your citations, so what they say about you is accurate and current.

You don't need to chase big-name publishers. In Parse's data, commercial media citations were spread across 73,928 domains, and the top 50 supplied only 26.3% (Parse). A niche trade blog that ChatGPT cites for your exact prompts is worth more than a famous outlet it never cites.

For a closer look at when roundup placements are useful and when they're just link building under a new name, see Are Third-Party Listicles Good GEO or Just AI-Era Link Building?

Step 6: Re-test and measure three signals separately

Most "it worked" claims blur three different signals. Track each one on its own:

Crawler visits need server-side logging. Browser analytics can't reliably see crawler requests. A user-agent string alone is also not proof of identity, because anyone can fake one. You can run curl -A "OAI-SearchBot" https://yoursite.com/ to confirm that your logging captures the request and your firewall doesn't block it. That test only checks your setup. A real visit needs verification against OpenAI's published IPs or a CDN's verified-bot flag (Tideflow AI).

Compare like with like. Re-test at least 24 hours after any robots.txt change. For content and PR work, re-run the full prompt set on a fixed schedule, weekly or monthly, rather than checking once. Improvement means the share of runs that mention or cite you went up across the set, not that one prompt changed once.

Doing this by hand across ChatGPT alone is manageable. Across ChatGPT, Gemini, Perplexity and Google's AI surfaces, with repeated runs and crawler logs, it becomes a part-time job. Tideflow AI is built for that tracking loop. It monitors those five engines and records mentioned, cited, rank and the sources each answer used. It logs AI crawler requests at the server or edge, marking OAI-SearchBot hits as verified only when Cloudflare's bot metadata confirms them, and it connects referrals to conversions. Tideflow is in early access and, like any tool, cannot guarantee that ChatGPT will recommend you.

What has no evidence behind it yet

Several tactics show up in most "get mentioned by ChatGPT" checklists without evidence that they affect ChatGPT citations. Treat them as optional experiments:

• Schema markup for ChatGPT citations specifically
• llms.txt files
• Freshness as a ranking factor, meaning "update content every X days"
• Guaranteed timelines such as "results in 60–90 days"

If you try one, change only that one thing between baseline and re-test so you can tell whether it did anything.

Frequently Asked Questions

Can I get ChatGPT to mention my brand without search, through training data?

Influencing ChatGPT's training-data knowledge of a brand is not something any reliable evidence explains how to do. Blocking or allowing GPTBot only controls whether your content may be used for training. It has no effect on ChatGPT search. Search is the path you can act on and measure.

Do the same steps work for Perplexity, Gemini and Google AI Mode?

The steps carry over, but the balance shifts by engine. In Parse's data, Google AI Mode drew far less recommendation evidence from brands' own sites (19.1%) than ChatGPT Search did (50.4%), and more from community and social sources (Parse). A brand with thin independent coverage is more exposed on Google's AI surfaces, and a brand with a weak website is more exposed on ChatGPT.

Is it legitimate to publish my own "best tools" list that includes my product?

Vendor-published roundups do get cited. About one in five vendor pages backing a ChatGPT recommendation had a roundup-style title (Parse). Such a list holds up only if it is honest: it names real alternatives, states who each option suits, and is clear about where your product is weaker. A self-serving list with no real comparison offers little that is worth citing.

Is answer engine optimization really different from SEO?

Much of the work overlaps with SEO, including crawl access, substantive pages and earned coverage. What changes is the measurement: prompts instead of keywords, citations instead of rankings, and repeated sampling instead of a single position. Is AEO Really a New Discipline? covers that debate in more depth.

Your first move this week

Check your robots.txt and firewall for OAI-SearchBot today. It's the one fix with an official, documented effect. While you wait the roughly 24 hours for it to apply, write your prompt set and record a baseline. Let the domains in that baseline decide whether your next month goes into owned pages or third-party coverage, then re-run the same prompts and judge the result by the share of runs that mention you, not by a single screenshot.

Sources

• OpenAI, Overview of OpenAI Crawlers; Publishers and Developers FAQ
• Capston, Earned Media Bias in AI Search: What the Toronto Study Reveals
• Parse, Who publishes the pages AI cites for buying advice?
• Aggarwal et al., GEO: Generative Engine Optimization
• GEO Wiki, GEO: Generative Engine Optimization (Aggarwal et al. 2024)
• Tideflow AI, Detect AI Bots; AI Visibility & Content Platform